Optimising ChatGPT for creativity in literary translation: A case study from English into Dutch, Chinese, Catalan and Spanish

2025-04-25

AI Summary

This study addresses the limited creativity of ChatGPT in literary translation from English into Dutch, Chinese, Catalan, and Spanish. We systematically investigate how prompt engineering, temperature settings (0.2–1.0), and text granularity (sentence- vs. paragraph-level) affect translational creativity. Methodologically, we introduce a novel multidimensional human evaluation framework and a quantitative creativity scoring formula. Results show that minimalist prompts (e.g., “Translate… creatively”) combined with temperature = 1.0 significantly enhance creativity—outperforming DeepL on Spanish, Dutch, and Chinese tasks. AI-generated translations achieve 70%–85% of the creativity level of professional human translations, yet remain systematically inferior to expert human performance. This work provides a reproducible methodology and empirical foundation for controllable creative optimization in LLM-based literary translation.

Technology Category

• Cognitive Modeling & Cognitive Systems: Computational Creativity
• Natural Language Processing: Prompt Engineering / Prompting
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

This study examines the variability of Chat-GPT machine translation (MT) outputs across six different configurations in four languages,with a focus on creativity in a literary text. We evaluate GPT translations in different text granularity levels, temperature settings and prompting strategies with a Creativity Score formula. We found that prompting ChatGPT with a minimal instruction yields the best creative translations, with"Translate the following text into [TG] creatively"at the temperature of 1.0 outperforming other configurations and DeepL in Spanish, Dutch, and Chinese. Nonetheless, ChatGPT consistently underperforms compared to human translation (HT).

Problem

• Evaluates ChatGPT's creative translation variability across languages
• Compares GPT configurations and DeepL for literary translation quality
• Assesses human vs ChatGPT performance in creative translation tasks

Innovation

• Optimize ChatGPT with varied configurations for creativity
• Evaluate translations using Creativity Score formula
• Minimal instruction prompts yield best creative translations